Mortality Burden Attributed to the Synergy Between Human Bio-Climate and Air Quality Extremes in a Climate Change Hotspot

Abstract

The Eastern Mediterranean is a rapidly warming climate change hotspot where heat and air pollution increasingly interact to affect human health. This study quantifies the mortality burden attributed to the synergistic effects of thermal stress and air pollution in Thessaloniki, Greece. Daily mortality data (2001–2019) were analyzed together with pollutant concentrations (PM10, NO₂, O₃) and the modified Physiologically Equivalent Temperature (mPET) using a hierarchical Generalized Additive Model with Distributed Lag Non-Linear terms to capture combined, lagged, and age-specific responses. A refined, count-independent definition of the Attributable Fraction (AF) was introduced to improve stability in small strata. The results show that heat and pollution act synergistically, explaining on average 20–30% of daily mortality during severe co-occurrence events. Seniors were most affected during hot, polluted summers (AF ≈ 27%), while adults showed higher burdens during cold, polluted winters (AF ≈ 30%). Intra-urban analyses revealed stronger simultaneous effects in the western, more industrial districts, reflecting combined environmental and socioeconomic vulnerability. The findings demonstrate that temperature extremes amplify pollution-related mortality and underline the need to integrate air quality and bioclimatic indicators into early warning and adaptation systems in Eastern Mediterranean cities.

1. Introduction

The Mediterranean basin, and particularly Southeastern Europe, is widely recognized as climate change hotspots, experiencing warming rates nearly twice the global average and a rapid increase in heat extremes [1,2]. In Greece, according to RCP8.5 projections, annual near-surface temperatures are expected to increase by roughly 1.6 °C in the near term and by up to 4.3 °C by late century [3]. This trend is coupled with a twelve-fold increase in the frequency of heatwave days in Thessaloniki [4], posing escalating risks to human health and urban resilience.

Heat-related health impacts are not solely driven by climatic factors but are increasingly shaped by demographic change. Population ageing is accelerating across Greece and the wider Southeastern Mediterranean, creating demographic conditions that heighten vulnerability to environmental stressors. Projections for Greece and Cyprus show a rapid increase in the proportion of adults aged ≥65 years due to low fertility, outward migration of younger groups, and rising life expectancy [5]. This demographic shift is closely linked to greater sensitivity to climate- and pollution-related risks, as older adults exhibit higher prevalence of chronic cardiopulmonary conditions and reduced physiological resilience. Recent public health assessments highlight that ageing populations in Greece will face disproportionate impacts from heat, cold, and poor air quality, particularly within a health system already under strain [6]. Evidence from a comprehensive time-series study on the island of Crete showed that older adults (≥65 years) experience significantly higher mortality during both heat and cold extremes, with temperature-related effects persisting across several lag days and disproportionately affecting cardiovascular and respiratory outcomes [7]. These findings highlight that ageing populations are inherently more vulnerable to thermal stress and underline the need to explicitly consider demographic structure when assessing climate-sensitive health risks in Mediterranean settings.

At the same time, exposure to air pollution remains one of the leading public health threats in Greece [8]. According to recent estimates, up to 13,000 premature deaths annually, approximately 10% of national mortality, may be attributable to PM2.5 and ozone exposure [9]. While national air quality levels are projected to modestly improve under mitigation scenarios, excess mortality is expected to rise due to ageing and urban demographic trends. Recent analysis for the Eastern Mediterranean region demonstrated that mortality increases during hot conditions, with significant short-term rises in cardiovascular mortality observed at temperatures exceeding local seasonal thresholds [10]. The authors also identified clear signs of maladaptation, reporting that heat-related mortality risks have intensified over time across several regions, including Athens and Thessaloniki. At the same time, a multi-country study has shown clear short-term increases in daily mortality associated with higher PM10 and ozone concentrations across Southern Europe, with estimated effects commonly falling within 1% and 4% per 10 μg/m³ increase [11].

Despite mounting evidence linking heat and air pollution to health outcomes, most studies assess these exposures in isolation [12,13,14]. However, the synergistic co-occurrence of thermal stress and poor air quality is increasingly recognized as a non-linear amplifier of health risk [15,16]. The modified Physiologically Equivalent Temperature (mPET) offers a thermo-physiologically consistent framework that better reflects the human body’s response to environmental stressors than air temperature alone [17,18]. Recent applications of mPET across demographic groups in Athens demonstrated its utility in quantifying demographic heat vulnerability trends, supporting both its scientific robustness and policy relevance [19], while its application in epidemiological studies has yielded more physiologically consistent and location-specific assessments of heat-related health risks, particularly when used in conjunction with hierarchical modelling frameworks [20].

To address these gaps, the present study examines the combined effects of thermal stress and air pollution on daily mortality in Thessaloniki, Greece, a rapidly warming urban area in the Eastern Mediterranean. Using a high-resolution bioclimatic indicator (mPET) together with pollutant concentrations (PM10, NO₂, O₃) and mortality records, we assess seasonal, demographic, and intra-urban variations in risk. We quantify the mortality burden attributable to non-optimal combinations of heat and air quality using an attributable fraction (AF) framework derived from the model-based risk estimates. Our aim is to provide an integrated understanding of how heat and air quality jointly influence mortality patterns in an ageing Mediterranean city.

2. Materials and Methods

2.1. Study Area

The Thessaloniki Urban Area (TUA), which consists of six municipalities, is located in northern Greece, positioned along the Thermaikos Gulf and forming the country’s second-largest metropolitan region with approximately 1 million residents according to the 2021 national census. Thessaloniki functions as a major economic centre in Southeastern Europe, with key activities including port-related logistics, manufacturing, higher education, and tourism. The city is situated within a coastal–continental transition zone and experiences hot, dry summers and mild, humid winters, characteristic of the Eastern Mediterranean climate [21]. Between 2001 and 2021, it has undergone a demographic shift toward an older population, with a 5% decline in the general population and a 45% rise in residents over 65 years old (Figures S1–S3), reflecting the broader national trend [22,23]. In recent years, the city has been increasingly affected by rising temperatures due to climate change, which exacerbates human health risks [10]. At the same time, air quality challenges remain persistent due to dense traffic, industrial activities in surrounding municipalities, and regional pollution transport from the Balkans and Central Europe [24,25,26]. These environmental and demographic characteristics render Thessaloniki an informative setting for assessing climate-sensitive health risks.

2.2. Mortality, Air Quality and Human Thermal Bioclimate Data

The mortality data used here were obtained by the Hellenic Statistical Authority. They span the period 2001–2019 (19 years) and include daily counts of deaths due to all natural causes at municipality level in TUA. The data were stratified by sex (male; female) and age (adults < 65 years old; seniors ≥ 65 years old). In total, 134,003 deaths are included in the analysis, of which ~78% (105,420) correspond to seniors, and ~14% (19,253) correspond to adults.

Hourly values of PM10, O₃, and NO₂ concentrations (all expressed in μg/m³) for the period 2001–2019 were obtained from three air quality monitoring stations, operated by the Ministry of the Environment and Energy. Main characteristics of AQ stations are presented in

Table 1. Metadata of air pollution monitoring stations of the National Air Pollution Monitoring Network, operated under the responsibility of the Ministry of Environment and Energy, used in this study.

Station Name	Longitude	Latitude	Altitude (m)	Measured Pollutants			Type of Stations
				NO2	O3	PM10
AUTH	22.955444° E	40.633143° N	29	YES	YES	YES	Urban Traffic
Agias Sofias	22.945099° E	40.633724° N	12	YES	YES	YES	Urban Traffic
Kordelio	22.893219° E	40.673453° N	30	YES	YES	YES	Urban Industrial

. For the analysis, the daily average PM10 and NO₂ values, and the daily mean 8 h moving average for O₃ over all stations were used. Further details on the seasonal distributions of NO₂, O₃ and PM10, together with the corresponding season-specific percentile values, are provided in Figure S4 and Table S1 of the Supplementary Material. For the calculation of AFs, pollutant levels were grouped into three exposure categories (low, medium, high), defined using the 5th, 50th, and 95th percentiles of each pollutant’s seasonal distribution.

The human biometeorological data used in the present study were obtained from the publicly available dataset of [27], which contains hourly values of population-weighted mPET index, computed using the RayMan Pro model [28], for diverse populations in Greece, spanning the period from 1991 to 2020. Importantly, mPET was not calculated from local monitoring station observations; instead, it was extracted directly from this national human biometeorological reanalysis, which is driven by high-resolution CERRA environmental fields and incorporates physiology-specific parameters for each population group. mPET offers a physiologically relevant estimation of heat stress by considering all factors influencing the human thermal environment. It is calculated using a semi-steady human energy balance model, which accounts for variations in anthropometric data, activity levels, and clothing across different population groups [17,18]. For this study, we used the daily max mPET values between 2001 and 2019 for the two targeted population classes (Adults, Seniors), following the methodology of [20]. Age-specific thresholds were derived using a representative reference individual for each group: a 35-year-old adult and a 70-year-old senior.

Traditionally, the thermal stress classification based on mPET is based on a nine-category assessment scale, spanning from extreme cold stress to extreme heat stress. Here, to enhance the clarity and interpretability of our results, we combined some of the mPET categories per season, as outlined in

Table 2. Adjusted mPET thresholds and thermal sensation classes used in the current study.

Thermal Sensation	Warm Period mPET Thresholds	Cold Period mPET Thresholds
“Cold”	-	<8 °C
“Cool”	<18 °C	8 °C–18 °C
“Comfortable”	18 °C–23 °C	18 °C–23 °C
“Warm”	23 °C–35 °C	>23 °C
“Hot”	>35 °C	-

.

For analytical purposes, the year was divided into a warm season (April–September) and a cold season (October–March), consistent with the general climatological characteristics of the Eastern Mediterranean region.

2.3. GAM Modelling Framework

In this study, we employed the statistical framework of penalized GAMs [29,30], which provides a flexible approach to modelling the heat–health nexus by accounting for non-linearity and lagged relationships, while objectively penalizing for over-fitting. This method allows for the exploration of complex interactions between exposures, such as heat and air quality, while also capturing variations across different districts and population subgroups [31]. The hierarchical structure of the GAM framework is particularly useful for overcoming challenges related to small population sizes and low death counts in the examined regions [20]. The model was implemented in R (version 4.4.2) using the mgcv package (version 1.9-1) [30], with restricted maximum likelihood (REML) estimation to prevent overfitting [29]. REML enables approximate Bayesian inference, allowing us to produce uncertainty intervals for each estimated quantity.

For each season s (cold/warm), region r (CE, N, W) and pollutant P (PM10, O₃, NO₂) a different model is implemented. The model is defined mathematically as follows:

$${{\rm N}}_{g,t}~ NegBin \left({\mu }_{g,t}, \theta \right)$$

(1)

$$\begin{array}{l}\log \left({\mu }_{g,t})\right)=a+{\displaystyle \sum _{l=0}^L}[h\left({mPET}_{t-l}, P_{t-l}, l\right)+h_g\left({mPET}_{t-l}, P_{t-l}, l\right)+{\beta }_{DoW\left(t\right)}\\ +f_{y\left(t\right),g}\left(t\right)+f_g\left(y\left(t\right)\right)]\end{array}$$

(2)

where ${{\rm N}}_{g,t}$ is the mortality count for each daily time step t and age group g (adults, seniors). The counts are assumed to follow a Negative Binomial (NegBin) distribution, which is appropriate for count data which may also exhibit overdispersion [31]. Parameter θ is a dispersion parameter that can allow for more variability than can be captured by the more traditional Poisson distribution. Then, μ_g,t is the mean mortality count for time t and age g, whose natural logarithm is modelled as an additive function of various “effects”. Firstly, h(mPET_t−l, P_t−l,l) is a 3-dimensional smooth function of mPET, pollutant and temporal lag. At lag 0 for instance, h(mPET_t,P_t,0) captures the joint effect of mPET and pollutant P on the number of deaths on day t, so that h(mPET_t−1,P_t−1,1) captures the effect of the day before and so on. The smooth function is constructed using a thin plate regression spline (TPRS), which is an optimal choice in the sense that it avoids knot selection and can accommodate many dimensions [30]. Then, the term $\sum _{l=0}^L\left[h\left({mPET}_{t-l}, P_{t-l}, l\right) \right]$ captures the aggregated synergistic effect of the two exposures across the temporal lags. Note that h(mPET_t−l,P_t−l,l) is common to both age groups, whereas the function h_g(mPET_t−l,P_t−l,1) (also constructed as a TPRS) is a “deviation” from h() for each group (again constructed using TPRS). This “main effect plus deviation” structure allows for pooling of risk information across the two age groups [31], while still enabling the estimation of a different risk profile for each.

Moreover, the model contains terms to capture confounding factors in terms of temporal structures. The quantity y(t) is a year counter and f_g(y(t)) is a smooth function (TPRS) of the yearly time step for each group, which captures the long-term trend in mortality (such as changes in population). Then, DoW(t) is the day of the week and the term β_DoW(t) is a random effect that captures the day-of-the-week effect in the reporting of mortality (weekends are slightly negatively biassed). Lastly, f_y(t),g(t) is a smooth function of the daily time step (TPRS), and there is one of these for each year and age group. These allow for day-to-day structured variability of mortality within each year and age group.

Implementation of the above model is performed in the R package mgcv [29,30,31], which uses Restricted Maximum Likelihood (REML) to estimate the unknown functions from the data. This implementation provides objective penalisation (regularization) in the estimates, meaning that the number of effective parameters (degrees of freedom) in each of the smooth functions is chosen by the data. The functions can for instance be reduced to linear if there is little evidence from the data that the relationship is non-linear. The only requirement is to ensure there are enough degrees of freedom to begin with, which can be assessed via the function k.check() in mgcv [31].

In addition to the city-wide model, we conducted a complementary intra-urban analysis to explore spatial variability in exposure–response relationships. For this purpose, the above model was applied separately to three sub-regions of the Thessaloniki Urban Area: Central-East (CE), North (N), and West (W). This stratification enabled us to examine whether combined heat–air pollution effects differed across areas with distinct demographic, land-use, and pollution characteristics.

2.4. Attributable Fraction

The term exp(α) captures the overall mean mortality count (for a specific region and season), since all terms in Equation (2) are centred on zero (they average to zero over the observed data—an identifiability constraint implemented by default in mgcv). As such, the term

$${RR}_g\left({mPET}_{t-l}, P_{t-l}, l\right)=\exp \left\{h\left({mPET}_{t-l}, P_{t-l}, l\right)+h_g\left({mPET}_{t-l}, P_{t-l}, l\right)\right\}$$

(3)

can be interpreted as the Relative Risk from mPET and pollutant at lag l, for the given age group g. This is the risk of a certain combination of mPET, pollutant and lag, relative to the average mortality in the data. The cumulative risk

$${CR}_g\left({mPET}_{t-l}, P_{t-l}, l\right)=\exp \left\{\sum _{l=0}^L\left[h\left({mPET}_{t-l}, P_{t-l}, l\right)+h_g\left({mPET}_{t-l}, P_{t-l}, l\right)\right]\right\}$$

(4)

is an aggregated risk measure that integrates the RR across the temporal lags. Then, the exposure values of mPET and the pollutant P that minimize the CR, can be thought of as the “optimal” mPET and P [32] defined here as T^(O) and P^(O), on the basis that any other exposure value will lead to an increase in risk. Then, the RR can be expressed relative to these optimal exposures rather than the mean mortality [31].

The (forward) AF [33] is a quantity that measures the proportion of mortality that can be attributed to non-optimal conditions. Specifically, for any given day t in the observed time series (with observed exposure values mPET_t and P_t), the AF is defined as:

$$AF\left({mPET}_t, P_t\right)=1- {\displaystyle \frac{CR({mPET}^{\left(0\right)}, P^{\left(o\right)})}{CR({mPET}_t,P_t)}}$$

(5)

This quantity, constrained between 0 and 1 (by definition of the optimum mPET_t and P_t) is interpreted as the fraction of deaths that are attributed to non-optimal mPET_t and P_t on a given day t. An overall measure of the AF across the observed time series is then defined as

$$\widehat{AF}={\displaystyle \frac{\sum _{t=1}^{n}AF\left({mPET}_t, P_t\right)\ast \widetilde{N_t} }{\sum _{t=1}^n{N}_t}}$$

(6)

where $\widetilde{N_t}$ is the mean number of deaths in the time interval [t, t + L] and $\sum _{t=1}^n{N}_t$ is the total number of deaths in the time series. In essence, for any day t, the attributable number of deaths is computed, then summed and scaled by the total number of deaths [33]. As such, $\widehat{AF}$ is a function of both the estimated relative risk but also the observed mortality data and despite its wide application, $\widehat{AF}$ is subject to sampling variability (noise) in the mortality counts. With low or zero death counts, $\widehat{AF}$ can be unstable as noise is introduced from the observed counts. Here, we alternatively propose to use as

$$\widetilde{AF}={\displaystyle \frac{1}{n}}\sum _{t=1}^{n}AF\left({mPET}_t, P_t\right)$$

(7)

instead of $\widehat{AF}.$ This is simply the mean attributable fraction from Equation (5) across the observed time series. It is a more robust metric, especially for low mortality counts and one that gives results that are comparable to $\widehat{AF}$. To evidence this claim, we present a comparison between $\widehat{AF}$ and $\widetilde{AF}$ using an open-source data set in the Appendix A.

3. Results

3.1. Observations on Population Dynamics, Mortality Trends, and Statistical Confidence

Census data from 2001 and 2021 reveal a steady population decline of −5% in the Thessaloniki Urban Area (Figure S1). However, despite the overall decline, the proportion of elderly residents has consistently increased by more than 45% (Figure S2). This demographic shift implies a growing vulnerability in the population, as older individuals are more susceptible to environmental and health-related stressors. This trend underscores the importance of accounting for age-related vulnerability when assessing mortality risks associated with air pollution and thermal stress.

From 2001 to 2019, the total number of deaths, aggregated across both warm and cold months, shows a clear upward trend (Figure S3). This increase further supports the notion of an increasingly vulnerable population and a growing public health burden. Consistently across all years, mortality during the cold season (70,031 across all years) exceeds that of the warm season (63,972), likely due to seasonal contributors such as respiratory infections, cardiovascular strain, and reduced indoor air quality. However, since approximately 2015, the gap between cold- and warm-season mortality appears to be narrowing. This may suggest rising exposure to heat-related health risks, including heatwaves and chronic disease exacerbation, potentially amplified by climate change.

95% confidence intervals and mortality risk significance for PM10 are displayed in Figure 1 (and Figures S5 and S6 for NO₂ and O₃, respectively). Seniors exhibit consistently higher confidence in RR estimates than adults. This can be attributed to both a higher absolute number of deaths in this age group and their increased physiological vulnerability to environmental stressors, which may strengthen the observed statistical associations. Across all three pollutants, the warm season is generally associated with greater risk than the cold season, particularly in seniors. Notably, regardless of age group, regions marked as “risky” (in red) are associated with 95% confidence, reinforcing the reliability of findings when significant associations are observed.

In contrast, adult-specific estimates during the cold season appear less robust, as reflected in the broader or weaker confidence intervals. Interestingly, grey areas indicating significant results are nearly absent in adult cold-season maps across all three pollutants. This may reflect the delayed physiological impact of cold exposure, which often extends over 2–3 weeks, whereas the analytical lags used here were shorter (7 days for PM10 and NO₂, 9 days for O₃). Furthermore, air pollution often exerts cumulative effects, possibly muting short-term associations. In seniors, however, the link between mortality and environmental risks remains strong enough to reach statistical significance even during the cold season, again underlining their heightened vulnerability.

3.2. Seasonal and Age-Specific Patterns of Attributable Mortality Fractions

The way we depict AF graphs (Figure 2 and Figure 3) provides a clear and informative visualization of combined effects of exposure to air pollution and thermal stress on mortality. The seasonal split is necessary, as the impacts of thermal stress and pollution vary significantly across warm and cold months. Age-stratified analysis reveals significant differences in vulnerability between adults and seniors.

PM10: The analysis reveals that both PM10 concentration levels and thermal stress significantly influence mortality, with notable differences across seasons and age groups. In the cold season, adults exhibit consistently high AFs across all PM10 levels, ranging 21.9–31.7%, with cold and warm thermal stress contributing substantially. Seniors, although displaying lower AFs overall in winter, show increasing vulnerability as PM10 levels rise, particularly under cold conditions (21.1%). In contrast, during the warm season, AFs increase with PM10, especially among seniors, where hot conditions become the dominant contributor under high pollution (26.9%). Adults also exhibit rising AFs in warmer months, though to a lesser extent, with cool conditions playing a major role at lower PM10 levels (22.7–26.4%).

NO₂: During the cold season, adults show consistently elevated AFs across all NO₂ levels, with cold stress contributing most substantially (23.2% at Low level), followed by cool and comfortable conditions. Interestingly, warm stress also plays a non-negligible role, especially under high NO₂ levels at almost 14%. In contrast, seniors exhibit lower overall AFs in winter, yet their vulnerability increases with rising NO₂ levels, particularly under cold conditions, where the attributable fraction becomes predominant. During the warm season, mortality risk appears to rise with NO₂ exposure, especially among seniors, where hot thermal stress dominates under high pollution, corresponding to an AF of 29.4%. Adults also exhibit increasing AFs with rising NO₂ levels in summer (≈12–20%), though the distribution is more balanced, with contributions from hot, warm, and even cool conditions.

O₃: Ozone-related mortality also exhibits distinct seasonal and age-dependent patterns. During the cold season, seniors face a substantially higher mortality burden compared to adults (at high level, 37.3% Cold stress for seniors compared to 29.4% for adults), with AFs rising consistently with O₃ levels. Cold thermal stress is the dominant contributor among seniors (between 21 and 37%), reflecting their heightened vulnerability to winter conditions. In contrast, adult AFs are lower and more stable, with a modest increase under cold exposure (up to ≈25%). In the warm season, adults exhibit significantly higher AFs than seniors, particularly under hot and warm thermal conditions (≈20–32%). The adult burden increases with O₃ levels, with heat-related stress accounting for over half the total AF at all pollution levels. Seniors show lower AFs during summer, with hot conditions contributing modestly (up to 19.3%). These patterns suggest that while cold-season O₃ exposure disproportionately affects the elderly, heat-related effects during the warm season primarily impact adults.

Across pollutants, mortality risk shows a robust seasonal signature: heat-related impacts dominate in summer, while cold-related ones prevail in winter. Temperature extremes consistently amplify pollution-related mortality, confirming the modifying role of thermal stress. Seniors are generally more affected during the warm season, adults during the cold season, except for ozone where this pattern reverses.

The observed differences in AFs can be explained by a combination of physiological, behavioural, and environmental factors. Adults’ higher respiratory rates and greater outdoor activity increase inhalation of pollutants [34,35], especially in Thessaloniki’s mild winters, leading to higher winter AFs for PM10 and NO₂. Conversely, seniors’ susceptibility rises in summer when heat stress exacerbates cardiovascular and respiratory fragility [36]. Despite spending more time indoors, older adults’ impaired thermoregulation and diminished adaptive capacity heighten their risk [37,38], particularly under high O₃.

3.3. Intra-Urban Differences in Attributable Mortality

To complement the city-wide findings, we also conducted a supplementary intra-urban analysis, focusing on three subregions of Thessaloniki: Central-East (CE), North (N), and West (W) (Figures S7–S9). While the statistical confidence at this level was generally limited, some indicative spatial patterns emerged that provide useful context.

In Central-East Thessaloniki (Figure S7), the most densely populated part of the city, cold-season AFs showed clearer gradients across all pollutant levels, particularly PM10, O₃, and NO₂. Cold and cool mPET categories dominated the burden, especially among older adults (16–74%), reflecting the seasonal and physiological sensitivities of this age group. Despite a 13.5% drop in overall population since 2001, CE has seen a 31.5% increase in its elderly residents, suggesting a shifting vulnerability profile (Figures S1 and S2).

North Thessaloniki displayed more complex and inconsistent patterns across both seasons (Figure S8). Trends in AFs were generally weaker and less structured, possibly reflecting the region’s smaller population size and increased uncertainty. Nonetheless, its demographic trajectory, featuring a modest total population growth (+2.4%) and a substantial 68% rise in elderly residents, indicates a potentially growing public health concern that remains partially hidden in the current results.

In Figure S9, West Thessaloniki, an area with a mixed industrial-residential character, showed more consistent signals. Extreme mPET categories were prominent across all pollutant levels and both seasons. Interestingly, seniors in this region exhibited lower cold-season AFs than adults, which may suggest behavioural differences, such as spending more time indoors. The broader demographic shift here is also noteworthy, with the total population growing by 14.9% and the elderly population nearly doubling (+83%) over the past two decades.

One interesting observation is that seniors, despite their greater physiological vulnerability, tend to exhibit lower AFs in some intra-urban settings. This may be explained by differences in lifestyle and exposure: older adults might be more cautious, less mobile, or more sheltered during environmental stress events [39,40], while working-age adults are more likely to be exposed to outdoor air pollution and temperature extremes during daily activities [41].

4. Discussion

To address the complex public health threat posed by thermal stress and air pollution in Mediterranean cities, this study focuses on quantifying their synergistic impact on mortality in Thessaloniki, Greece. Building on our previous work [42], we introduce a refined, physiologically consistent framework that integrates mPET with a hierarchical Generalized Additive Model (GAM) and Distributed Lag Non-Linear Models (DLNMs), tailored to capture seasonal, demographic, and intra-urban differences.

A key innovation of this study lies in its application of an extended definition of Attributable Fraction (AF) [33]: one that does not rely on observed daily mortality counts, but instead uses a counterfactual baseline derived from the model predictions under optimal exposure conditions. This approach provides greater robustness in the presence of low or zero mortality counts, which are common in stratified data, and allows for clearer spatial and seasonal attribution. By integrating mPET and pollutant concentrations (PM10, NO₂, and O₃) within a synergy-aware AF framework, we are able to identify critical exposure combinations and high-risk population segments.

Ultimately, the goal is to produce policy-relevant, spatially disaggregated estimates of mortality burden, highlighting when and where risk converges across environmental and demographic dimensions. This approach provides a scalable foundation for climate health early warning systems and evidence-based adaptation planning, especially in ageing and environmentally burdened urban areas.

In particular, by integrating the modified Physiologically Equivalent Temperature (mPET) with daily concentrations of PM10, NO₂, and O₃, we quantified the age-, season-, and region-specific mortality burden in Thessaloniki between 2001 and 2019. The results reveal that (i) mortality risk increases under concurrent high temperature and pollution levels, with AFs reaching 26–29% under extreme summer PM10 and NO₂ conditions and up to 37% under winter O₃ peaks; (ii) vulnerability varies systematically by age and season, with seniors most affected during hot and polluted summer periods (AF ≈ 27%) and adults during cool polluted winters (AF ≈ 30%), and (iii) intra-urban patterns reflect both demographic ageing and socioeconomic heterogeneity across the city.

These findings are consistent with an expanding body of literature demonstrating that the effects of temperature on mortality are affected by air pollution in a non-linear way and thus cannot be assumed separable. A global systematic review [43] concluded that co-exposure to heat and air pollutants, particularly ozone and particulate matter, produces synergistic rather than additive health effects. At the regional scale, [44] showed that in California, days with concurrent extreme heat and PM2.5 increased mortality by 21%, well above single-exposure estimates. Likewise, [11] analyzed 620 cities worldwide and found consistent effect modification of temperature–mortality relationships by ambient pollution. Recent works [45,46] further support these observations, indicating that high pollution amplifies the slope of heat-mortality curves by up to 15% and that the magnitude of synergy depends on urban morphology and greenness. Within this context, our findings extend the empirical basis for the South-Eastern Europe by providing physiologically consistent, quantified evidence of such interacting risks.

The amplification of mortality under combined stressors can be attributed to well-established physiological and chemical mechanisms. At the organism level, co-exposure strains thermoregulatory and cardiovascular systems, promoting oxidative stress and impaired gas exchange [36,37]. The strong summer signal in seniors observed here (AF ≈ 27% for high PM10 level and mPET > 35 °C) aligns with these mechanisms, as ageing reduces sweat response, cardiac reserve, and vascular plasticity. Similar patterns have been observed across European cities [38], where older adults experience disproportionate increases in hospitalisations and mortality during joint heat–pollution episodes.

Conversely, adults under cold, polluted conditions displayed higher AFs than seniors reaching 31% for PM10 and 23% for NO₂, a reversal consistent with behavioural exposure patterns. Working-age adults spend longer periods outdoors, commute through traffic, and engage in moderate activity that increases minute ventilation. Evidence [39,41] confirms that daily mobility and time–location patterns substantially shape exposure, often offsetting physiological resilience. In contrast, older adults’ lower AFs in some intra-urban settings (<20%) may reflect self-protective behaviour (e.g., greater caution, limited mobility, and heat-avoidance strategies) [38,40] which reduce direct exposure even though their intrinsic susceptibility remains high. These interactions between physiology and behaviour underscore why age alone cannot fully predict vulnerability without considering exposure context.

The intra-urban gradients observed within Thessaloniki mirror broader environmental-justice patterns reported internationally. However, statistical confidence at this spatial scale was generally lower due to the smaller population sizes within each sub-region, which increased uncertainty in AF estimates. In the Central-East district, where population density is highest and the share of residents ≥65 years increased by 31.5% despite an overall 13.5% population decline, cold-season AFs for seniors reached up to 40% under O₃ extremes. Western Thessaloniki, a mixed industrial-residential zone that saw 83% growth in its elderly population, exhibited pronounced hot-season AFs (>70%) and stronger responses to combined heat–pollution exposure. Northern Thessaloniki showed weaker but more variable trends, consistent with its smaller population base and lower data confidence. Such spatial contrasts likely reflect both micro-environmental variability and social inequalities. Internationally, [47] showed that low-income and minority U.S. communities face systematically higher PM2.5 and NO₂ exposure; [48] confirmed similar patterns across European cities, and [49] linked urban poverty globally with proximity to emission sources. Thessaloniki’s demographic and land use gradients thus provide a microcosm of these global inequities, highlighting the intersection of environmental and social vulnerability.

From a methodological perspective, the study contributes to advancing joint-risk assessment by employing an extended, count-independent definition of the Attributable Fraction, improving stability where daily deaths are few. This modification reduced variance in low-count strata compared with the classical AF definition (Appendix A). The combination of mPET with pollutant concentrations in a hierarchical GAM–DLNM framework offers a reproducible approach for other Mediterranean and European urban areas seeking to quantify combined exposures with demographic granularity.

The implications for policy and practice are substantial. Treating heat and air pollution as independent hazards likely under-estimates mortality during concurring events. Integrating the two within early warning systems (for instance, issuing joint heat–pollution alerts) could therefore enhance prevention. The distinct seasonal and demographic patterns observed here call for targeted interventions: heat-health action plans prioritizing older populations during summer pollution peaks, and occupational or commuting protection measures for working-age adults during winter pollution episodes. The spatial heterogeneity across Thessaloniki further underscores the need for place-based mitigation, such as expanding urban greenery, improving ventilation corridors, and reducing emissions in densely populated western districts. These strategies align with equity-focused frameworks in European adaptation policy, which advocate explicit attention to socially vulnerable urban populations.

Despite its strengths, several limitations must be acknowledged. Reliance on fixed-site monitoring may under-represent personal exposure, especially indoors. Mortality data, aggregated at municipal level, could mask finer social gradients. Although mPET provides a more realistic indicator of human thermal bioclimate than air temperature alone, residual confounding from humidity or unmeasured co-pollutants cannot be excluded. Future work should incorporate high-resolution exposure models, wearable sensor data, and socioeconomic indicators to clarify mechanisms driving intra-urban disparities. Longitudinal analyses could also explore how adaptation measures or behavioural change modify combined risk relationships over time.

In summary, our quantified assessment demonstrates that combined heat and air pollution exposures can account on average for 20–30% of total daily mortality during severe co-occurrence events. The Mediterranean basin, characterized by rapid warming, persistent air quality challenges, and accelerating population ageing, stands at the frontline of this joint hazard. Quantifying and understanding these interactions are essential for developing equitable, evidence-based public health strategies capable of safeguarding vulnerable communities in a changing climate.

5. Conclusions

The combined heat–pollution burden can account for a substantial share of daily mortality, typically around 20–30% during severe co-occurrence events, with even higher values in specific seasons and sub-regions, particularly among older populations. The results confirm that temperature extremes amplify pollution-related mortality, with distinct seasonal and demographic patterns with seniors being more affected in summer, adults in winter. Although intra-urban estimates carry lower confidence due to smaller populations, they reveal clear spatial disparities linked to ageing and socioeconomic factors. In the context of the Eastern Mediterranean climate change hotspot, these findings highlight that combined environmental stressors are becoming major drivers of health risk. Integrating air quality and human thermal bioclimatic indicators into regional early warning tools and adaptation frameworks is therefore essential to enhance climate resilience and public health preparedness across vulnerable urban populations.